Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.793663
Title: Deep learning-based food image classification and crowdsourcing-based calorie estimation approach to support dietary management
Author: McAllister, Patrick
ISNI:       0000 0004 8503 6271
Awarding Body: Ulster University
Current Institution: Ulster University
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Food logging is a technique that is used to monitor nutritional intake and research states that food logging is essential for weight management. The aim of this research was two fold; to investigate, develop, and evaluate computer vision and deep learning approaches for food image classification and also to explore and evaluate methods that could be used for predicting nutritional content in food images. In regards to food image logging, this research applied computer vision and deep learning techniques to detect and classify food items in photographs and utilised crowdsourcing along with image processing methods to estimate calories of food portions for dietary management. This research presents studies that inform the development of an automated food image logging platform that combines image classification and calorie estimation to support dietary management. This thesis consists of four studies, Chapter 3 and Chapter 4 research, develop, and evaluate image based approaches and crowdsourcing for calorie estimation. Chapter 5 and Chapter 6 focus on developing and evaluating approaches for food image classification. Chapter 5 investigates the use of conventional image feature extraction approaches with supervised machine learning classification algorithms in classifying a range of food image datasets. Chapter 6 investigates and evaluates the use of deep residual convolutional neural network (CNN) features in classifying a variety of food image datasets. Chapter 7 synthesises results from each study and discusses how they can be integrated into a dietary management framework. This research contributed to the literature by proposing a dietary management system that combines deep learning approaches for food image classification and crowdsourcing for calorie estimation. A novel crowdsourcing calorie adjustment approach was proposed to promote accuracy in food logging for dietary management along with combining state-of-the-art ResNet-152 CNN deep feature extraction with machine learning models to classify variety of food image datasets.
Supervisor: Zheng, Huiru (Jane) ; Moorhead, Anne ; Bond, Raymond Sponsor: DEL
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.793663  DOI: Not available
Keywords: Obesity ; Calorie ; Deep learning ; Image classification ; Machine learning ; Classify ; CNN ; Features ; Crowdsourcing ; Image
Share: